Energy and Climate Economics - Munich, 4-5 March 2021 Carbon Pricing of International Transport Fuels: Impacts on Carbon Emissions and Trade ...
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Energy and Climate Economics Munich, 4–5 March 2021 Carbon Pricing of International Transport Fuels: Impacts on Carbon Emissions and Trade Activity B. Gabriela Mundaca, Jon Strand, and Ian R. Young
Carbon pricing of international transport fuels:
Impacts on carbon emissions and trade activity*
By
B. Gabriela Mundaca
World Bank
bgmundaca@gmail.com
Jon Strand
World Bank
jstrand1344@gmail.com
Ian R. Young
Department of Infrastructure Engineering
University of Melbourne
ian.young@unimelb.edu.au
March 2, 2021
JEL codes: CO2 emissions; maritime trade; carbon taxation; core competence
Key words: F14 F18 F64 H23 H87 Q54 Q56
Abstract
We study impacts of carbon taxation to international transport fuels on CO2 emissions and trade
activity, focusing on maritime transport which constitutes the most important international trade
transport activity for the years 2009-2017. The bunker-price elasticities range from -0.003 to -0.42.
For the current level of international trade, a global tax of $40 per ton CO2 tax, will reduce carbon
emissions by about 7% for the heaviest traded products (at the 6-digit HS levels of aggregation)
transported by sea. The greatest CO2 emission reductions are for products with particularly low
value-to-weight ratios such as fossil fuels and ores.
_______
* We are grateful for the comments received during the Carbon Pricing Leadership Coalition (CPLC) Research
Conference in New Delhi, India, February 14 – 15, 2019; the January 2020 American Economic Association meetings
in San Diego, CA, USA, and the University Rochester’s Virtual International Trade & Macro Seminar. We are also
very thankful for suggestions and detailed comments from Jonas Teusch, and Kurt Van Dender. We are the only
responsible for the contents of this paper. Any opinions and conclusions expressed herein are those of the authors and
do not necessarily represent the views of the World Bank Group. All errors are our own.1. Introduction
To limit drastic climate change and its devastating consequences, it is necessary to implement
appropriate and optimal policy instruments in core economic sectors to reduce greenhouse gas
(GHG) emissions at a global scale. No international transport activity today faces any meaningful
tax for the CO2 it emits. This has at least three adverse consequences for the shipping sector, of
main concern of this paper. The first is a higher than optimal activity in international shipping
(types of vessels and their technologies, their travel routes, and the amounts and types of goods
being transported), as the sector does not face the true global costs of international trade. The
second is too high fuel consumption (and too polluting fuels), and consequently too high CO2
emissions. The third is low fiscal revenue raised from international shipping transport, a common
and critical problem for many low-income countries with low tax revenues collection (see Keen,
Parry, and Strand (2013) for further arguments). Today, the shares of global CO2 emissions due to
international aviation and shipping are each about 2.5%. According to Cristea et al. (2013), 51%
of CO2 emissions from international trade in 2004 resulted from sea freight; 27% from air freight;
and 22% from land (road and rail) transport.
This paper aims to contribute to better understanding of how and by how much CO2 emissions
from maritime transport can be reduced by implementing a carbon tax. As far as we are aware, this
is the first theoretical and empirical study to analyze the impacts of changes in bunker fuel prices,
which proxy carbon taxation, on global international trade and on global CO2 emissions resulting
from this trade, using a comprehensive panel dataset for products at the 6-digit HS level of
aggregation covering the years from 2009 to 2017. UNCTAD (2009) indicates that fuel costs
account for 50% to 60% of total ship operating costs depending on the type of ship and service
(see also Gohari et al. (2018)). Since carbon taxes are not today in place, we use changes in the
bunker price to simulate the effects of a possible carbon tax in the maritime sector.
This work seeks to provide a guidance to the international community about how to attribute
responsibility per country/region and traded product type to their shares in the global CO2
emissions in the maritime sector, and how carbon taxation could affect CO2 emissions. We
estimate CO2 emissions levels from maritime transport of products at the 6-digit HS level of
aggregation.
This study considers all possible worldwide country pairs that trade the heaviest products (6-
digit HS level of aggregation). It focuses on two main topics. First, we analyze theoretically and
1empirically the impact of changes in bunker prices on trade (intensive margin). Second, we
estimate the possible impact on carbon emissions as a result of changes in the structure of the
international goods’ trade that follow from carbon taxation (i.e. inferred from changes in bunker
prices), for 2030 and 2050.
In principle, carbon pricing (which includes carbon taxation) could impact on carbon emissions
from international goods freight in three main ways: 1) via changes between and within the three
main modes of transport, sea, air and land transport; 2) changes in the structure of trade including
the weight of shipped products and the choice of trading partners and product types; and 3) changes
in energy use per ton-km by transport mode. For international goods transport, sea transport
dominates, but all three modes are important. Apart from land-based transport, international person
transport is dominated by aviation. While people-oriented transport represents 85% of the aviation
sector’s revenues (although a lower share of ton-km), 90% of international sea transport’s revenues
are derived from goods transport.
There are two main alternatives for implementing a carbon price for goods transport: i) carbon
taxation (a given tax per unit of carbon emissions); and ii) cap-and-trade schemes for trading
rights to emit carbon at a (positive) carbon price established in the carbon market. In both cases, a
carbon price represents the marginal cost of carbon emissions related to bunker fuel consumption
by the maritime sector. Under carbon taxation, substantial revenues can be raised, some of which
can be transferred to the poorest and most remote countries with high and increased trade costs
(which could lead to fewer product varieties, and lower traded quantities); and/or to support global
climate finance purposes. Offset or other cap-and-trade schemes are less likely to raise similar
revenues. It is important to emphasize that an efficient strategy to implement a carbon tax to bunker
fuel requires this tax to be universal, worldwide cooperation, and avoidance of free riding problems
in order to prevent the carbon leakage phenomenon (Elliott et al. 2010).
In our study we consider carbon taxation, but our results and conclusions will hold if carbon
pricing is implemented through a cap-and-trade or offset scheme (given a positive and reasonably
stable global carbon price for international transport fuels), instead of through a carbon tax. Note
also that our analysis sheds light on how carbon emissions can gradually be reduced when
imposing a carbon tax on bunker fuels. The crucial issue we here consider is: how to immediately
contribute to reducing the current, highly excessive, GHG emissions rate toward long-run
sustainable levels.
2GHG emissions from international transport have recently become a central issue of interest,
for various reasons. The adverse consequences mentioned above are increasingly recognized by
more countries and other international stakeholders. In 2017, the IMO implemented a new set of
technical carbon intensity vessel standards, and in April 2018, the International Maritime
Organization (IMO) decided on a plan to reduce the GHG emissions from international shipping
to half the 2008 level (1,135 million tons) by 2050 (see IMO (2018)). However, without
implementing any type of carbon pricing, the Third IMO GHG Study (IMO (2015)) projects that
conditional on future economic and energy developments, and planned vessel efficiency
improvements, maritime emissions could increase by between 50% and 250% by 2050 under a
business-as-usual scenario. All these projections under different scenarios are greater than
emissions for 2012. International shipping emissions continue to be omitted from the GHG
emissions accounts of the countries involved, as they are only referred as supplementary
information in national inventories for communication to the UNFCCC (Nunes et al. (2017)).
Due to lack of data, we do not study here how carbon taxation could: i) reallocate trade of
products between transport modes (air and sea transport); ii) induce more technologically efficient
and less carbon intensive shipping modes; and iii) incentivize development of more
environmentally friendly fuels.
In the continuation we present a literature review in Section 2, while the theoretical model is
presented in section 3; the data description and the empirical analysis and results of the effect of a
carbon tax on international trade can be found in section 4. Section 5 presents the estimations of
the potential reductions in CO2 emissions that could result from implementing carbon taxation to
shipping international trade. Section 6 includes the possible financial resources that can be
obtained from carbon taxation at the international level. Section 7 sums up and concludes.
2. Review of related literature
The background literature dealing directly with the main research topics of our paper is limited.
We here revise the studies closest to this paper. Cristea et al. (2013) computed GHG emissions
from both production and transport (air, rail and trucks) of internationally traded goods for one
year, 2004, using data from the Global Trade Analysis Project (GTAP). They considered 28
countries and 12 (own-defined) regions; and 23 traded goods sectors and 6 non-traded service
sectors. We instead consider worldwide trade between country pairs of all the possible heaviest
3products at the 6-digit HS level of aggregation. Their paper did not study either the impacts of fuel
price changes (or carbon taxation) on international trade, which is here our main objective.
Shapiro (2016), using a gravity model, estimates the effect of transport costs on trade values,
for US and Australian imports over the period 1991–2010, for 13 sectors; but does not estimate
CO2 emissions. These are obtained from separate sources: from production using GTAP for 2007,
and for airborne and maritime trade from the International Air Transport Association (IATA) and
IMO, respectively. Shapiro considers a single emissions intensity rate of 9.53 grams CO2/ton-km
for the entire maritime sector, to estimate the effect on welfare of a carbon tax. We instead take
into account that carbon emission intensity rates vary by ship type, and that goods are transported
in different types of ships according to the product type. His paper does not present the impacts of
the counterfactual carbon tax on CO2 emissions, as we analyze.
Parry et al. (2018) consider that a carbon tax of $75 per ton CO2 by 2030 and to $150 per ton
in 2040 on international shipping, could affect: 1) ships’ technical efficiency improvements; 2)
ships’ operational efficiency improvements; 3) optimal ton-kilometers of trade transport activity;
and 4) traded volume in ton-kilometers. These factors together will contribute to a reduction in
CO2 by 14% by 2030 and 23% by 2040, respectively. A reduction in trade would only contribute
with 4% of such CO2 reductions.
Brancaccio et al. (2018, 2020) analyze, among other things, the effects of shipping costs and
oil prices on only bulk shipping using two databases, a small sample of shipping contracts from
Clarkson; and the Automatic Identification System (AIS) data on ship movements. Brancacchio et
al. (2020) find that a 10% increase in the oil price, and in total shipping costs, reduces bulk shipping
by 4.4%, and 10%, respectively. Their trade elasticity with respect to the oil price varies between
-0.1 (when the oil price is low) and -1.2 (with high oil price). These effects are larger than those
found in our study. Both these papers rely on only bulk shipping, and a limited sample of shipping
contracts. We use global maritime trade data, and a wider range of products at the 6-digit HS level
of aggregation. Neither of these two papers studies impacts on carbon emissions, which is our
objective.
Two recent papers consider impacts on global GDP levels due to carbon taxes on shipping.
Lee et al. (2013) study impacts of different fuel tax levels charged to container ships, using the
GTAP-E model, and find negligible impacts on the global economy for low carbon tax levels, but
more significant impacts if the tax is US$90/ton of CO2, with the greatest relative impacts on
4China. Certain distant trade routes are discouraged by high carbon taxes. Sheng et al. (2018)
consider more modest carbon tax (US$10–25/t CO2), using a global recursive dynamic CGE
model, and find that global GDP is likely to be reduced by 0.02 – 0.05%. Trade weights and
patterns are affected, but only moderately.
Limão and Venables (2001) and Behar and Venables (2011) studied the effect of transport
costs on volume of bilateral trade using gravity models. They do not analyze specifically the effects
of fuel prices, but find that trade volumes decrease as transportation costs rises.
A strand of literature also estimates the spatial and temporal variability of GHG emissions from
shipping traffic. Wang et al. (2007) use the waterway network ship traffic, energy, and
environment model STEEM to quantify and geographically represent inter-port vessel traffic and
emissions. Their model also estimates energy use, and assesses environmental impacts of shipping.
The area of study is North America: United States, Canada, and Mexico, for 2002. The Third GHG
Study of IMO (IMO (2015)), also using AIS data, presents a detailed and comprehensive global
inventory of shipping emissions, but in somewhat less detail for spatial and temporal variability of
global emissions than Johansson et al. (2017), who use STEAM3 and the AIS data to estimate
global shipping emissions for 2015. Schim et al. (2018), using data from AIS and trade custom
declarations, calculate carbon emissions per vessel and per journey for Brazilian export shipments
in 2014. This literature does not address carbon taxation nor its possible effects on international
maritime trade, as we do here.
Thus, in comparison with the related existing literature, we analyze econometrically how CO2
emissions can be reduced by implementing carbon taxation on international maritime trade. We
consider worldwide country pairs and all traded heaviest products to obtain the effects of carbon
taxation on trade and CO2 emissions by product type. We also take into consideration that emission
intensities vary substantially by vessel type and the type of product vessels transport. We also
analyze theoretically how international trade (intensive margin) is impacted by carbon taxation.
3. The theoretical model
3.1 Background
Our key analytical framework is based on recent international trade theory and serves as the
basis for our econometric analysis of the possible impact of carbon tax per ton of fuel on the
intensive margin of international trade, the choice of trading partners (and implicitly the distance
the products will travel), and on carbon emissions.
5We will not focus on the networking between firms on both sides of a trade transaction
(Bernard and Moxnes (2018), Bernard et al. (2018), Bernard et al. (2017)), as complete data for
all firms participating in international trade in all countries are not available to us.
One of the main distinctions between our work here and the gravity models is that we consider
the effect of carbon taxation on the combination of the quantity of trade of products at the 6-digit
HS levels times the distance the exported product travels, and not aggregate flows of trade at the
country level. Thus, exporters choose not only export quantities but also the distance to their
importing countries to minimize costs, including transportation costs. We have also one practical
reason for estimating the elasticity of weight – country distance with respect to bunker fuel price
is because to calculate CO2 from maritime trade. We need to take into consideration that carbon
emissions intensities by type of ship and products it transports, are measured in ton-kilometers.
Calculation of impacts of carbon taxation on carbon emissions from maritime transport then
requires us to obtain the elasticities of ton-kilometers with respect to the fuel price for each 6-digit
HS product category. We nonetheless consider several of the variables that are usually used for
the estimation of gravity models. We again remark that we only consider the heaviest products,
i.e. products that have been consistently the heaviest in each of the years between 2009 and 2017.
Even though our analysis and empirical implementation uses country data instead of firm data,
our theoretical model follows a bottom-up analysis from (exporting) firms’ behavior to countries’
determination of trade of products at the 6-digit HS level of aggregation.
We follow closely the theoretical underpinnings of activities of multi-product firms in
international trade (see Bernard, Redding, and Schott (2010, 2011); Arkolakis and Muendler
(2010); Eckel and Neary (2010); Mayer et al. (2014); and Eckel et al. (2015)). In our framework,
consumers in the importing countries choose which and how much of each product variety to
import.
In contrast to Armington (1969) and Shapiro (2016) who assume that each country produces
only one variety and that varieties are different across countries; we consider that each country
produces and exports a set of product varieties. This approach will help us to determine i) what
product varieties (at the 6-digit level) that are traded between the different country pairs could be
most affected by the implementation of a carbon tax; and ii) which of these products are the highest
emitters of CO2. This approach is crucial in order to attribute as correctly as possible the
responsibility of CO2 emissions by industry, and product type.
6Bernard, Redding, and Schott (2010, 2011) pioneered the modelling of asymmetries between
products from the demand side. In their work, firms consider their productivity levels and product–
market–specific demand shocks, before deciding to enter international markets. A firm then
determines the scale and scope of sales in different markets, and leads to a negative correlation
between prices and output prototypes. On the other hand, Eckel and Neary (2010) consider
asymmetries between products on the cost side (of producing different varieties), and find that
price and output prototypes are always positively correlated. We here integrate demand and supply
approaches by assuming that the marginal costs of producing a variety of products and fuel costs
determine the scale and scope of international trade, including the total distances that product
varieties travel, which implicitly implies choosing the trading partners.
Our main contribution to the theoretical literature is to consider a typical consumer in an
importing country as maximizing a two-level utility function that depends on the consumer’s
consumption levels (weight) of product varieties, from different industries. We model the typical
exporting multi-product firm in any given industry making decisions about i) the scale and scope
of its product varieties taking into account the marginal costs of each of its multi-products; and 2)
the importing countries (i.e. distance) where the products will be sold. We follow the approach of
Eckel and Neary (2010) and Mayer et al. (2014) by considering that firms that produce several
product varieties, face “product ladder” costs. Each firm then has a core product (its “core
competence"), with lower efficiency (higher costs) for products further away from this core.
We here define the core competence in which the cost linkages are both across product varieties
and trading partners. Thus, an exporting firm’s decisions about the weights and the distances that
any of its product variety travels, depend on the marginal cost of both producing such a product
variety, and delivering it to the specific importing country.
We do not consider a general equilibrium model as in Eckel and Neary (2010), to analyze
factor markets as an important channel for transmission of external shocks. Our available data does
not allow us to ascertain how factor prices and employment at our product level of disaggregation
will be affected by changes in fuel prices and determine the general-equilibrium adjustments. Thus,
we focus on a partial-equilibrium model (and reduced-form) analysis of how bunker price changes
(or carbon taxes) affect international trade and CO2 emissions of different products at the 6-digit
HS levels of disaggregation.
7To sum up, our theoretical model considers the impacts of changes in the bunker costs per ton-
km per type of product and vessel on: i) the traded weight – country distance of different 6-digit
products; and ii) CO2 emissions after taken into consideration that ships specialize in transporting
different products and have different carbon emission intensities.
3.2 The model
3.2.1 DEMAND
On the demand side, we follow closely Eckel and Neary (2010), in which consumers in the
importing countries buy different product varieties i from a total of Nj varieties in each industry j.
There are m importing countries and k exporting countries. The typical consumer in each of the m
importing countries maximizes a two-level utility function by choosing a level of consumption
qm(i;j) of the product variety i produced in industry j. We thus define the product variety i ∈[1,Nj],
where Nj is the measure of product variety i in industry j; and j changes over each interval [0,1].
At the lower level, the typical consumer in a given importing country m has a quadratic utility
function that depends on all the varieties this consumer chooses from industry j, defined by:
Nj Nj
N j
2
1
u[qm (0; j ),..., qm ( N =
j ; j )] am ∫ qm (i, j )di − bm (1 − ξ m ) ∫ qm (i, j ) di + ξ m ∫ qm (i, j )di ; (1)
2
0
2 0 0
Nj
where ∫ qm (i, j )di in equation (1) is the consumption of all varieties in a given industry j by this
0
typical consumer in the importing country m. The utility parameters am, bm and ξm are assumed to
be non-negative. They denote the consumers’ maximum willingness to pay, the inverse market
size, and the inverse degree of product differentiation, respectively. If ξm =1, the goods are
homogeneous (perfect substitutes), so that demand only depends on aggregate industry output. ξ=0
describes the case when the demand for each good is completely independent of other goods. Thus,
the last two terms in equation (1) indicate that consumers give increasing weight to the distribution
of consumption levels across varieties.
The upper utility levels for this typical consumer in each of our m importing countries are
defined by an additive function of a continuum of quadratic sub-utility functions, where each sub-
utility u[qm(0;j)} …, qm(Nj;j)] (as defined in equation (1)) corresponds to industry j:
1
U [(u{qm (0;1),..., qm ( N1 ;1)}),..., (u{qm (0; j ),..., qm ( N j ; j )})] = ∫ u{q
j =0
m (0; j ),..., qm ( N j ; j )}dj. (2)
8The problem for the typical consumer in each importing country is to maximize a two-tier
utility function with respect to q(i,j):
N j 1
),..., qm ( N j ; j )})] ∫ am ∫ qm (i, j )di −
U [(u{qm (0;1),..., qm ( N1 ;1)}),..., (u{qm (0; j=
j =0 0
; (3)
Nj
Nj
2
1
bm (1 − ξ m ) ∫ qm (i, j ) 2 di + ξ m ∫ qm (i, j )di dj
2 0 0
subject to the following budget constraint:
1 Nj
∫ ∫ p(i, j ) * q
0 0
m (i, j )di dj ≤ E ; (4)
where p(i,j) is the (global) price of product variety i from industry j 1; and E denotes the expenditure
by the typical consumer of an importing country on a set of differentiated products from different
industries in different exporting countries.
To solve this optimization problem (equations (3) and (4)), we use the Lagrange multiplier
method with λ as the Lagrangian multiplier (i.e. marginal utility of income). We assume that the
budget constraint (4) is binding and solve the following Lagrangian maximization problem:
1
max
qm ( i , j ), λ ∫ (u{q
j =0
m (0;1),..., qm ( N1 ;1)}),..., (u{qm (0; j ),..., qm ( N j ; j )})dj −
(5)
1 Nj
λ ∫ ∫ p (i, j ) * qm (i, j )di dj − Em
j =0 0
after considering equations (3) and (4).
The typical consumer’s optimal choice of variety i's amount (in any given industry j) is given
by the first-order condition after maximizing (5) with respect to demand of any variety i at any
given industry j:
N
am − bm (1 − ξ m )qm (i ) + ξ m ∫ qm (i )di − λ p (i ) =
0. (6)
0
The individual (inverse) linear demand function for product variety i is:
N
am − bm (1 − ξ m )qm (i ) + ξ m ∫ qm (i )di .
λ p(i ) = (7)
0
1
Good markets are fully integrated (Eckel and Neary (2010)).
9We can assume, without losing generality, that the budget constraint is always binding and
λ=1. We also consider that consumers are identical within any importing country and across m
importing countries. Thus, the utility parameters am, bm and ξm are assumed to be identical for all
consumers of variety i from any given industry j, which allows us to drop the subscript m.
To move from individual to aggregate demands, we assume that, the price of a variety i is
the same everywhere; and there are L consumers (with identical preferences) in each of k importing
countries. Thus, the aggregate demand for variety i in country m, will be equal to q(i,m)×L. Thus,
the inverse of the global demand for product variety i in any given industry j, that each exporting
firm will face, as:
N
a b (1 − ξ ) ∫ q (i, m)dm + ξ ∫ ∫ {q (i, m) × L}dmdi ;
p (i ) =− (8)
m 0 m
where ∫ {q (i, m) × L}dm is the global demand for product variety i.
m
3.2.2 SUPPLY
The typical firm z may export the product variety i to only M countries, thus M ≤ m countries. In
addition, the exporting firm z in any given industry and country k, produces a number of product
varieties equal to ρz, where 0 < ρz < Nj. These are to be exported to a portfolio of importing countries
M located at specific distances (δMz) from the exporting firm location. For any given level of the
bunker fuel prices, exporters choose where to export (i.e., the size of M), which implicitly allows
them to minimize the total distance δMz their products need to travel, and minimize costs.
If we normalize the number of consumers L to be equal to 1, we define each firm as maximizing
the following profit function:
ρz
π z (k )
= ∫ ∫ { p (i)} − C (i, M ;δ
M 0
z z Mk ) * qz (i, M ) di dM ; (9)
where p(i) is the price of product variate i in any given industry j; Cz(i,M;δmk) comprises the
following costs: i) shipping costs (insurance rates, tariffs, border and port prices, bunker oil price,
the type of ship used to transport the product variety i, and the distance between the exporter’s
country and the importer’s country (δMk)) per unit of the exported product variety i from firm z
(located in exporting country k) to importing country M. For given geographical locations of the
importing countries, fuel prices are expected to have substantial influence on the exporter’s cost;
10and ii) the marginal cost that exporting firm z faces to produce variety i, and is related to its core-
competitiveness of producing a specific product variety i (in any give industry j). Marginal cost
increases as the exported product variety moves away from the “core competence” of the exporting
firm at which its marginal cost is lowest. Indeed, this “core-competence” here plays a crucial role
for the net effect that the bunker price per traded ton-km can have on the structure of trade and
finally on carbon emissions.
We solve for the amount of product variety i to be exported by the typical firm z (in any
given industry j and country k) to country M, qz(i,M), when it faces the global demand
∫ {q(i, m)}dm for variety i. We shall assume that firms (like firm z) play a single-Cournot game,
m
choosing simultaneously the number of product varieties (ρz), the quantity of each variety to
produce and export, and the set of importing countries (and implicitly δMz), taking into account
that rival firms do not change their scale or scope. Keep in mind that the global price for product
variety i will be p(i). Furthermore, qz(i,M) > 0 when its marginal profit of producing variety at the
equilibrium price p(i) is also greater than zero, and this demanded variety i is within the firm z’s
varieties scope of production (ρz) and countries scope (M). Hence, we can substitute the limit to ρz
for Nz and M for m, add a constraint that qz(i,M) > 0 only if it is optimal for firm z to produce that
variety and export to the specific importing country, otherwise qz(i,M) = 0.
Substituting equation (8) into equation (9), and taking the first-order condition for the typical
firm z in any given industry, we obtain:
∂π z ' a − 2b(1 − ξ )q (i ', m) − 2bξ qz ' (i ', m)
0≤ = z'
∂qz ' (i ', m) λ
N z Nz
b (1 − ξ ) ∫ ∫ qz (i, m)dmdz + ξ ∫ ∫ qz (i, m)didm + ∫ ∫ ∫ qz (i, m)dmdzdi
− z≠z ' m m i ≠ i ' m z ≠ z ' i ≠i '
; (10)
λ
−C z (i, M ; δ Mk ) ⊥ qz ' (i ', m)
∂π z '
where the operator ┴ ensures that either = 0 (i.e. the firm z’ is producing a variety i
∂qz ' (i ', m)
optimally) or qz’(i',m) = 0 (i.e. the firm z’ does not produce variety i). This complementary
condition accounts for firm z’ selecting the number of varieties (ρz) to produce. Setting equation
(10) equal to zero and solving for qz'(i',m) we obtain:
11qz ' (i ', m) =
z
N Nz
a − b (1 − ξ ) ∫ ∫ qz (i, m)dmdz + ξ ∫ ∫ qz (i, m)didm + ∫ ∫ ∫ qz (i, m)didzdm
z≠z ' m m i ≠ i ' m z ≠ z ' i ≠i ' C z (i, M ; δ Mk )
−
2b 2b
(11)
In equation (11):
a. The aggregate output for the typical firm z is:
N j ,z
=Qz ∫∫
m i ≠i '
qz (i, j )didm + ∫ q (i ', j , z ', m)dm
m
b. The total global demand for product variety i in any industry j is:
=D(i, j ) ∫ ∫q
z≠z ' m
z' (i, j )dmdz + ∫ q (i ', j , z ', m)dm
m
c. The global demand for all the product varieties in industry j is:
Nj
=F ( j) ∫ ∫ ∫q
m z ≠ z ' i ≠i '
z' (i, j )didzdm + ∫ q (i ', j , z ', m)dm
m
Thus, equation (11) shows the unique Cournot solution (for nonnegative values of a and b)
with multiple firms. It further indicates that the amount of product variety i qz’(i,m), that firm z will
export depends on the different costs its total aggregate output, the total world production of variety
i, and global production of all varieties that are part of industry j.
4. The econometric analysis and the data
4.1 The empirical strategy
We use the System of Generalized Methods of Moments (GMM) [Arellano–Bover
(1995)/Blundell–Bond (1998)] for panel data as our estimation method. Our econometric strategy
is to instrument for the exchange rate, the bunker price per ton of fuel and the marginal costs of
producing product variety i. An ideal instrumental variable is one that is highly correlated with
these three variables but not with unobserved shocks to traded weight (quantity equation) of the
traded products. However, it is challenging to find the most appropriate and effective instrumental
variables. We have chosen as instruments: i) number of terror attacks to oil field; ii) level of trade
backhaul multiplied by the distance between trading partners; and iii) average wind speed and
12wave heights in the travelling routes between country pairs trading products internationally using
maritime transport. 2
Note that we take into account the theoretical foundations of the System GMM, which are to
use lagged variables of the model (except the dependent variable) as instruments for the equation
in first differences; and lagged variables in differences as instruments for the equation in levels.
We test the validity of the instruments with the Sargan test. When our econometric relation
includes the bunker price per metric ton of fuel, no time-fixed effect will be included to avoid
collinearity problems.
We also obtained the two-step estimates which yield theoretically robust results (Roodman
(2009)). Note also that, by applying the two-step estimator, we can obtain a robust Sargan test
(same as a robust Hansen J-test). This is important for testing the validity of the instruments (or
overidentifying restrictions). The validity of the model depends also on testing the presence of
first- and, in particular, second-order autocorrelation in the error term, as explained by De Hoyos
and Sarafidis (2006).
4.2 Data
As mentioned, our empirical analysis uses country data instead of firm data taking into
consideration that our theoretical model follows a bottom-up analysis from (exporting) firms’
behavior to countries’ determination of trade of products at the 6-digit HS level of aggregation.
Our data source is World Integrated Trade Solution (WITS) from the World Bank which
contains bilateral international trade in terms of weight and value by product and year, at the 6-
digit HS levels of aggregation. Our original dataset consists of approximately 3.9 million
observations for the period 2009–2017, including worldwide trading country partners, and more
than 6,000 commodities at the 6-digit HS level of aggregation. We only consider the products that
have had the highest weight consistently each year during our period of study. These products
make up more than 75% of the total weight of internationally traded goods transported by sea, and
are thus highly significant in terms of their total fuel consumption, and total carbon emissions from
international maritime trade. Our chosen 6-digit HS products belong to 21 (2-digit) industries.
2
We think that these instruments are relevant and appropriate given the recent work of Baumeister and Hamilton (2019) who
have concluded that supply shocks, such as geopolitical variables mentioned above, have been more important in accounting for
historical oil price movements than was found before in previous studies such as the work of Kilian and Murphy (2012, 2014).
13After we excluded all the landlocked countries and close land connected neighbors, we ended up
with around 2.2 million observations.
We also use the data from the Centre D’Études Prospectives et D’Informations Internationales
(CEPII) called GeoDist. This dataset has an exhaustive set of gravity variables developed in Mayer
and Zignago (2005) that allows us to analyze market access difficulties in global and regional trade
flows. GeoDist can be found online (http://www.cepii.fr/anglaisgraph/bdd/distances.htm) for
empirical economic research including geographical elements and variables. A common use of
these files is the estimation by trade economists of gravity equations describing bilateral patterns
of trade flows as functions of geographical distance.
Bunker price changes are here interpreted as proxies for changes in bunker fuel taxes. The
bunker fuel price data (in $ per metric ton) are available for the period between 2009 and 2017.
Relevant macro data at the country level from the World Development Indicators from the World
Bank have been used. The data for fuel (bunker) consumption by vessel type for ships come from
the ITF/OECD; see ITF (2018).
The data for terrorism events come from the Global Terrorism Database (GTD (2019))
developed by the National Consortium for the Study of Terrorism and Responses to Terrorism
(START) at the University of Maryland (2019). The source for backhaul trade is UNCTAD (2018)
(https://unctadstat.unctad.org/wds/TableViewer/tableView.aspx?ReportId=32363), and for wind
speed and wave height is Ribal and Young (2019).
4.3 The econometric model for estimating the effect of bunker price changes on the weight-
distance for the heaviest products at the 6-digit HS level of aggregation
Our empirical specification is tied closely to our theoretical modeling. Since each firm chooses
not only how much of each variety to produce but also to which countries to export, we study
econometrically the impacts of fuel price changes on the weight times shipping distance of traded
goods (in ton-kilometers). The distance is determined by the location of the importing and
exporting country.
Again, we do not have firm level data, only data for the traded quantity of products at the 6-
digit HS levels of disaggregation between country pairs (exporting countries versus importing
countries). We shall then interpret each data point on the traded quantity of product variety (6-digit
HS level) between two countries as the aggregate demand from the importing country satisfied by
14the supply of the exporting country’s aggregate decisions of its firms about exporting product
variety i, and behaving as Cournot competitors when deciding on exporting product variety i in.
When we consider the bunker price per ton of fuel, our proposed econometric model for the
bilateral trade between a pair of countries for a product variety at the 6-digit HS level of
aggregation is represented by the following empirical relation:
α11 + β11 ln( Bun ker pricet ) + λ11 ln( Exchange Ratet ) + ς 11Ckt +
ln qijkmt =
(12)
a11Qit + b11 Dijt + c11 Fjt + γ 11M mt + δ11 X kt + µijkm + ϕijkmt .
Note again that we do not include time-fixed effects to avoid collinearity problems with the
bunker price. In equation (12), at time t, qijkmt is the weight-distance measure and is obtained by
multiplying the weight of product variety of type i (i.e. a 6-digit product) from the j industry, traded
between the importing country m and the exporting country k at time t, times the distance between
country m and country k. φjkmt is a random disturbance term; while μijkm is product/industry –
importing/exporting effects. The variable definitions are given in Table 1.
Table 1. Definition of variables
EXPLANATORY DEFINITIONS
VARIABLES
qijkmt Weight of product of variety i (i.e. a product at the 6-digit HS level of aggregation) from
industry j (i.e. 2-digit industry) traded between the importing country m and the
exporting country k in time period t, times the distance between country m and country
k.
Xkt The exporting country k’s characteristics in year t: GDP growth rate, level of GDP in
US$, Inflation rate, population, 1st official language, if a colonizer, if a colony, Current
Account/GDP, and other variables considered in gravity modelling
Mmt The importing country m’s characteristics in year t: GDP growth rate, level of GDP in
US$, Inflation rate, population, 1st official language, if a colonizer, if a colony, Current
Account/GDP, and other variables considered in gravity modelling.
Ckt The (log) of sales value of a 6-digit HS level product, traded between two countries. The
higher its value, the closer is the product to the core competence of the exporting
country.
Price (pijkmt) (log) Total value of the 6-digit HS level products divided by total weight of the 6-digit
HS level products (within each 2- and 4-digit category, respectively).
Qit Total output of variety i in exporting country j
Dijt Total world output of product variety i
Fjt Global output of all the varieties in industry j
The marginal cost Ckt of a traded product, according to the theory, decreases as the product
variety (to be exported) moves closer to the “core competence” of the exporting firm at which its
marginal cost is lowest. We do not have data on the marginal cost that a typical exporting country
k’s incurs to produce variety i. Therefore, in this study, the marginal cost of producing variety i is
represented by the position of this variety i in terms of its sales value, when compared to all
15varieties sold by country k in each year t to its country partners (importers). Thus, for the triplet
exporter-product variety-importer by year, the product variety with the lowest sales value is the
lowest ranked (rank=n) product in the exporting country’s product portfolio; the product variety
with second lowest sales value is the second lowest core product (rank=n-1); and so on. A similar
approach has been used by Chatterjee, Dix-Carneiro, and Vichyanond (2013). The parameters β11
and ξ11 should both be negative, according to our theory.
Once we determined the marginal cost Ckt of the traded products, we have further defined the
sales value ranking (core-competence) into 4 categories:: category 1 contains the exported
products closest to the core competence of the exporting country, beyond the third quartile of the
country’s sales value; category 2 corresponds to products between the third quartile and the median
of the sales value; category 3 is the group with products between the median and the first quartile
of the sales value; and category 4 comprises the products furthest away from this core competence
or below the first quartile of the sales value. Thus, products in category 1 are the ones with the
lowest marginal costs, Ckt. We estimated equation (12) for each of these 4 core-competence
ranking categories for our heaviest 6-digit HS level products. From comparing the 4 estimates for
β11, for example, we learn whether and how the effect of changes in bunker fuel price on the
weight-distance of traded product i varies according to the exporting country’s marginal cost of
producing and exporting this product i.
We here focus on explaining the empirical results of the impact of annual changes in the global
average bunker price on the weight-distance of the heaviest products at the 6-digit HS level of
aggregation, traded bilaterally. These estimated elasticities, reported in Table 2, are crucial to
predict the possible reductions in carbon emissions due to a carbon tax. We again interpret changes
in the bunker prices as a measure of changes in a carbon tax.
The elasticities of the weight – distance with respect to the bunker fuel price across our 4
categories (core-competence ranking categories) can vary (for 6-digit HS products) from –0.003
(in the furniture industry) and -0.022 (in the automobile industry), to -0.4 (for ores) and -0.42 (for
fossil fuels). The elasticities of traded weight-distance with respect to the bunker price thus vary
greatly by industry. Figure 1 illustrates these differences when considering the average elasticities
across different core competence measures of the products.
16Table 2. The effect on trade weight-distance of changes in bunker prices. Heaviest Products at the 6-digit HS level of
aggregation. GMM estimates. Core linkages across products and trading partners (Standard errors in parentheses)
Industry Category
of the 6-digit HS products
10: Cereals 12: Misc.grains 15: Animal- 23: Animal 25: Salt, 26: Ores 27: Fossil 28:
(soya, etc.) Vegetable oils fodder stones, fuels Inorganic
cement chemicals
(1) (2) (3) (4) (5) (6) (7) (8)
Elasticity
lnBunkerPrice
Category 1 -.3167253 -.267283 -.1194384 -.226084 -.168464 -.399043 -.268136 -.185988
(.0384785) (.038564) (.0360417) (.025037) (.036067) (.040514) (.099553) (.014499)
Category 2 -.2708739 -.204212 -.1335528 -.180023 -.075225 -.364205 -.419043 -.119009
(.050876) (.05023) (.0303404) (.023125) (.028269) (.036674) (.048573) (.029796)
Category 3 -.1342549 -.182484 -.1195597 -.171604 -.046518 -.381997 -.399921 -.055966
(.0613848) (.05887) (.0273407) (.023708) (.045944) (.054714) (.034374) (.031242)
Category 4 -.1766457 -.240899 -.1629229 -.203508 -.078417 -.309274 -.419406 -.088050
(.0341268) (.032778) (.031513) (.027254) (.028332) (.041377) (.029511) (.029196)
29: Organic 31: Fertilizers 38: Other 39: Plastics 44: Wood 47: Wood 48: Paper 72: Iron &
chemicals chemicals pulp Steel
Elasticity
lnBunkerPrice
Category 1 -.2503799 -.3136591 -.133128 -.171858 -.126398 -.120166 -.081344 -.314474
(.011859) (.023212) (.02378) (.012968) (.015270) (.042681) (.015465) (.015874)
Category 2 -.1747001 -.184081 -.077378 -.143714 -.150651 -.166612 -.108389 -.185106
(.019566) (.054366) (.034452) (.025138) (.018592) (.017303) (.021645) (.021639)
Category 3 -.1446763 -.1415578 -.049072 -.132576 -.118661 -.128015 -.086957 -.189936
(.040146) (.048201) (.030936) (.020052) (.014973) (.023359) (.023483) (.027626)
Category 4 -.308295 -.1286671 -.1002387 -.181394 -.119221 -.099471 -.134247 -.318016
(.012813) (.040055) (.035148) (.022008) (.018626) (.022824) (.018002) (.043459)Industry Category
of the 6-digit HS products
73: Iron & 74: Cooper 76: Aluminum 87: Vehicles 94: Furniture
steel products
(17) (18) (19) (20) (21)
Elasticity
lnBunkerPrice
Category 1 -.185685 -.181851 -.125604 -.086893 -.086946
(.028941) (.039596) (.027870) (.027381) (.021733)
Category 2 -.090727 -.063762 -.104594 -.071957 -.071329
(.022915) (.070681) (.028389) (.042104) (.025531)
Category 3 -.011024 -.211283 -.103306 -.021779 -.044445
(.042178) (.031674) (.041742) (.022268) (.036509)
Category 4 -.003480 -.137916 -.077377 -.045349 -.003279
(.028299) (.026170) (.028155) (.021938) (.024746)
18Figures 1 and 2. Elasticities of weight-distance to changes in bunker prices for the
heaviest products at the 6-digit HS level of aggregation by industry type
Figure 1 Figure 2
Average elasticities and the elasticities to core
products (i.e. highest sales values)
Industry Category for Heaviest 6-digit HS products
FossilFuels(27) Ores(26)
Ores(26) Cereals(10)
Iron-steel(72) Iron-steel(72)
Cereals(10) Fertilizers(31)
Grains(soya)(12) FossilFuels(27)
Org.chem.(29) Grains,seeds(12)
Anim.Fodder(23) Organic Chem.(29)
Fertilizers(31) Animal fodder(23)
Inorganic Chem.(28)
Plastics(39) Iron-steel(73)
Copper(74) Copper(74)
(Animal-VegOils)15) Plastics(39)
Wood(44) Cement(25)
WoodPulp(47) Other chem.(38)
Inorg. Chem(28) Wood(44)
Paper(48) Aluminium(76)
Aluminium(76) Wood pulp(47)
Cement(25) Animal-Veget.oils(15)
Misc.Chemicals(38) Furniture(94)
Iron-steel manuf.(73) Vehicles(87)
Paper(48)
Vehicles(87)
Furniture(94)
-.4 -.3 -.2 -.1 0
-.4 -.3 -.2 -.1 0 Elasticity of High Sales Value Products Average Elasticities across Sales Value Products
Average Elasticities across Sales Values
Since the heaviest goods categories by 6-digit sectors considered by us constitute almost 75%
of total traded weight, this implies a very substantial impact of fuel taxation on fuel consumption
and carbon emissions for maritime trade of these heaviest products, as shown in the next section.
Note also that the elasticities of traded weight-distance with respect to the bunker price vary
greatly not only with the industry the 6-digit products belong to, but also with 1) the core
competence of the traded good, the closer is the product to the core competence of the country’s
product portfolio, the lower the net elasticity is; and ii) the higher value of the traded product
relative to its weight is, the lower the elasticity is. Figure 2 illustrates the average elasticities of
our heaviest products at the 6-digit HS level by industry category, and the corresponding
elasticities for the high-core products (highest sales values). Figure 3 shows the elasticities and the
US dollars (real of 2010) per kg of traded product by industry category.
Regarding the elasticity of weight – distance with respect to the marginal cost, we have found
that the further away a traded product variety moves from the core competence of its country’s
product portfolio, the larger the decrease in the traded weight – distance will be for that product.Figure 3. Elasticities of weight-distance to changes in bunker prices for the heaviest
products at the 6-digit HS level of aggregation by industry type; and average trade values
(in 2010 US$)
Cement(25)
Ores(26)
Fossilfuels(27)
Cereals(10)
Fertilizers(31)
WoodPulp(47)
Wood(44)
AnimalFodder(23)
InorganicChem(28)
Iron-steel(72)
Grains(12)
Animal-Veget.Oils(15)
Paper(48)
Iron-steel manuf.(73)
OrganicChem(29)
OtherChem(38)
Plastics(39)
Furniture(94)
Aluminium(76)
Copper(74)
Vehicles(87)
-.5 -.2 0 2 4 6 8
(mean) Elasticity (mean) Traded (2010) dollars per kg
Due to space constraints, we do not present in the paper the effects of other background
variables such as the exchange rate, Mnt, Xkt, Qit, Dijt, Fjt, the marginal costs, and the main statistical
tests (Sargan and autocorrelation), but they are available upon request. We however confirm that
for example a depreciation of the importing country’s currency reduces the weight-distance of
traded goods; while higher population in the importing country increases the weight-distance of
trade products.
5. Estimation of changes in carbon emissions due to carbon taxation
The CO2 emissions, and changes in such emissions as a result of increases in carbon prices, depend
on the type of product and the types of vessels with which the different products are transported.
To estimate CO2 emissions, we use data on fuel CO2 intensity per ton-kilometer (i.e. grams CO2
per ton-km) for 8 types of vessels and the types of products they transport for international trade.
These data come from the International Transport Forum (ITF) at the OECD (ITF (2018)). See
Table 3. The ITF/OECD provides this carbon intensity index for every 5 years, with historical data
since 2000, and projected figures up to 2050. These estimated emissions rates vary over time
20primarily according to the projected technological progress in the shipping industry, but also vessel
size and speed among other characteristics for each ship category. There are also data on carbon
intensities by vessel size for each vessel type. We here focus on the average size per vessel type,
to estimate average emission rates by vessel type as shown in Table 3, as we do not have data on
which product varieties are transported by which ship sizes.
5.1 Methodology to calculate carbon emissions
Table 3 shows that CO2 emissions rates by vessel type, in grams per ton-km of freighted goods by
2030, vary substantially from a low value of 3.7 grams for oil tankers, to a high value of 33 grams
for vehicle carriers. Assuming a single emissions rate for all ship types (as in Shapiro 2016) will
then lead to very large errors when calculating the carbon emissions for particular goods
categories, which we here avoid. We use Table 3 to estimate: i) the average annual CO2 emissions
between 2009 and 2017 (Business as Usual (BAU) CO2 emissions) for our set of heavy 6-digit
traded products; and ii) the average change in CO2 emission that would have taken place if a carbon
tax would have been implemented during this period (BAU emissions). It was then crucial that we
consider the ton-km (and not just tons) from trading the heaviest (6-digit HS) products between all
possible country pairs, to estimate the elasticity of this traded ton-km with respect to the bunker
price per ton. 3 These elasticities vary according to the core competence of this traded product, and
we take this into account when estimating the total carbon emissions by product and in aggregate.
Table 3: Average freight emissions by vessel type and transported product type
Type of ship Types of goods transported Carbon Intensity
(= grams CO2/ton-km
2010 2015 2030 2050
Bulk carriers Bulk agriculture, forestry, mining, minerals, non- 4.79 4.63 4.17 3.63
ferrous metals, coal products
Container ships Processed food, textiles, wearing apparel, leather 19.56 18.9 17.03 14.83
products, wood products, paper, iron and steel,
transport equipment, electronic equipment,
machinery and equipment, other manufactures
General cargo Food products, fish, livestock 13.88 13.41 12.09 10.52
Oil tankers Oil 4.32 4.17 3.76 3.27
LNG ships Gas 14.37 13.88 12.51 11.27
Products tankers Petroleum 14.0 13.53 12.19 10.62
Chemical ships Chemical products 10.29 9.94 8.96 7.8
Vehicle carriers Vehicles (automobiles) 37.92 36.63 33.01 28.74
Source: International Transport Forum (ITF, OECD) (2018)
3
The weight of the exported commodity is measured in tons and the distance in kilometers.
21To calculate the CO2 emissions for any given product type, we proceed as follows:
a. Using the information in Table 3, we multiply the amount of the CO2 or carbon intensity (in
grams of CO2) per ton-km emitted by the vessel type transporting the specific product type,
times the ton-km for the traded product type.
b. One ton of bunker fuel consumption corresponds to emitting 3.11 tons of CO2 (Olmer et al.
(2017)), which is used to calculate how the bunker fuel price per ton will change when
introducing a carbon tax. For every US$ of carbon tax per ton of CO2, the bunker fuel price
will increase by US$3.11. A carbon tax of US$ 40 per ton of CO2 would increase the bunker
fuel price per ton of fuel by US$124.4 (= 3.11 (carbon content of 1 ton of bunker fuel) times
$40). 4 If the bunker price is $450 per ton of fuel, the increase in the bunker price will be
124.4/450 = 0.27644 (or 27.64%) due to the US$ 40 of carbon tax per ton of fuel consumed.
c. We do not have data on the fuel consumption for each our traded 6-digit HS products. To be
able to use procedure (b), we assume that our bunker price elasticity with respect to the traded
ton-km equals the bunker price elasticity with respect to bunker fuel consumption.
d. The reduction in CO2 emissions for any product type, due to a US$ 40 carbon tax, is estimated
by multiplying the following components: i) the bunker price elasticity with respect to the
traded ton-km for the specific product type ((i.e. β11 in equation (12)); ii) the relative increase
in the bunker price, as calculated in (b) (e.g. 0.2764); iii) the total CO2 emissions for the given
ton-km of the traded product type as indicated in (a).
e. The estimations in this section considers the core competence with cost linkages across both
product varieties and trading partners. The Appendix presents an alternative set of estimations,
based on the core competence model with cost linkages only across product varieties.
5.2 Carbon emissions in the maritime sector for heaviest products: BAU and with carbon tax
Table 4 presents our bottom-up carbon emissions calculations by industry. These estimates take
into account: i) the average annual weight–distance (ton-km) of our heaviest 6-digit products; ii)
the average carbon intensities for the type of ship used to transport the different product categories
(see column 3 in Table 4), for our period of study (2009 – 2017) 5; iii) the elasticities of the ton-km
4
Note that US$40 is a lower bound of a range for the optimal global carbon tax (US$ 40 – US$ 80 per ton of CO2) to
be implemented from 2020 (Stern, Stiglitz and others (2017)).
5
We assume that the technological progress that took place in each of the ship type between 2009 and 2017, and
consequently their emission intensities, has been the same as the technological progress that occurred between
2010 and 2015.
22of traded 6-digit products between two countries with respect to the bunker fuel price; iv) that
these elasticities vary according to the core competence of the traded 6-digit product; and v) that
one ton of bunker fuel consumption corresponds to emitting 3.11 tons of CO2.
Recall that the “BAU” activity level for each sector corresponds to the average annual
activity levels between 2009 and 2017. We find that the (BAU) average annual carbon emissions
from transporting our heaviest products at the 6-digit HS level of aggregation (belonging to 21
industries) were about 450 million tons of CO2 (see column 4). This estimate is about half of total
CO2 emissions from the entire shipping sector over the same period (see e.g. IMO (2015)).
We also estimate what would have been the annual average reductions in CO2 emissions
(BAU emissions), and percentage reductions, if a global carbon tax of $40 per ton CO2 on bunker
fuel for transporting heavy products, would have been implemented each year from 2009 to 2017.
See respectively columns 5 and 6 in Table 4. Column 6 shows a reduction in total CO2 emissions
by about 6.6% relative to the BAU emissions. There are however substantial differences in
reductions by sector. By far the greatest reduction in tons is estimated to take place for the freight
of fossil fuels (by oil tankers), which also have the highest emissions of CO2. Emissions from
transport of fossil fuels can be reduced by around 12 million tons (or about 8%) due to this carbon
tax. The largest percentage reduction is however for ores (10.9%) followed by cereals (8.4%),
fossil fuels (8%), and iron and steel (7.8%). These are all very reasonable results as these goods
categories are all heavy relative to their values, and fuel costs should have significant impact on
the amounts of these goods transported. See Figure 4.
23You can also read
